# Subagent Orchestrator

> Shen-Hua orchestration workflow for coordinating multiple Codex subagents or simulated specialist roles. Use when the user explicitly asks for subagents, parallel agents, delegation, or multi-agent work, or when a complex task can be split into independent explorer, architect, worker, tester, reviewer, or docs-researcher streams under the current runtime policy.

- **Type:** Skill
- **Install:** `agentstack add skill-fengsihao23-eng-shen-hua-subagent-orchestrator`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [fengsihao23-eng](https://agentstack.voostack.com/s/fengsihao23-eng)
- **Installs:** 0
- **Category:** [Agent Skills](https://agentstack.voostack.com/c/agent-skills)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [fengsihao23-eng](https://github.com/fengsihao23-eng)
- **Source:** https://github.com/fengsihao23-eng/shen-hua/tree/main/skills/subagent-orchestrator

## Install

```sh
agentstack add skill-fengsihao23-eng-shen-hua-subagent-orchestrator
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# Subagent Orchestrator

Subagent Orchestrator splits complex work into bounded specialist tasks while keeping one main agent responsible for integration and final correctness.

## Trigger Conditions

Use this skill when:
- The user asks for subagents, custom agents, delegation, parallel work, or multi-agent execution.
- `$project-commander` identifies independent work streams that can run in parallel.
- Research, implementation, tests, and review can be separated without overlapping write sets.

## Do Not Trigger

Do not use this skill when:
- The runtime has no subagent tool and simulated roles add no value.
- The next critical-path step is blocked on the delegated work.
- Tasks would duplicate each other or write to the same files without a merge plan.
- Current tool policy requires explicit user permission and the user has not provided it.

## Role Catalog

- `explorer`: answer a specific codebase question; read-only.
- `architect`: assess design options and risks; usually read-only.
- `worker`: implement a bounded change with explicit file/module ownership.
- `tester`: run focused verification and report failures.
- `reviewer`: inspect diffs for bugs, regressions, and missing tests.
- `docs-researcher`: check external or local docs when current information may have changed.

## Orchestration Workflow

1. State the local critical path: what the main agent will do immediately.
2. Split only independent sidecar tasks.
3. For each subagent task, specify:
   - role
   - objective
   - allowed files/modules
   - forbidden files/modules
   - expected output
   - whether edits are allowed
4. Spawn real subagents only if available and permitted. Otherwise, run the role sequence locally.
5. Keep working locally while subagents run.
6. Wait only when the result is needed for the next step.
7. Integrate outputs. Review any subagent code before trusting it.
8. Update `.agent/` memory with assignments, results, and integration status.

## Required `.agent/` Updates

Always update:
- `.agent/TASKS.md`: add each delegated task and owner/role.
- `.agent/LOG.md`: record spawned or simulated agents, prompts, results, and integration notes.
- `.agent/HANDOFF.md`: list active/completed subagent streams and unresolved dependencies.

Update when relevant:
- `.agent/DECISIONS.md`: delegation strategy, accepted design choice, or merge decision.
- `.agent/PLAN.md`: phase changes caused by findings.

## Other Skills

- Use `$deep-planner` before orchestration if the task is not yet decomposed.
- Use `$verification-loop` after integrating any edits.
- Use `$self-reviewer` after integration.
- Use `$handoff-writer` before ending.

## Verification

Subagents do not replace verification. After integration, the main agent must run or follow `$verification-loop`.

## Output Format

```markdown
ORCHESTRATION PLAN
Main-agent critical path:
Delegated streams:
- Role:
  Objective:
  Scope:
  Output:
  Status:
Integration plan:
Verification plan:
```

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [fengsihao23-eng](https://github.com/fengsihao23-eng)
- **Source:** [fengsihao23-eng/shen-hua](https://github.com/fengsihao23-eng/shen-hua)
- **License:** MIT

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-fengsihao23-eng-shen-hua-subagent-orchestrator
- Seller: https://agentstack.voostack.com/s/fengsihao23-eng
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.